SEO Content Ops With Claude: Briefs, Refreshes, Links

#01Where does Claude fit in an SEO workflow?

Most SEO content work follows the same chain: pick topics from keyword and performance data, research what a good answer looks like, write a brief, draft, edit, publish, then refresh over time. Claude helps at almost every step, but it doesn't replace two things.

First, it doesn't have keyword volumes, rankings or click data built in. Those come from Google Search Console, your analytics and whatever SEO tool your team uses. You can paste exports in, upload them as files or connect tools through connectors where they exist.

Second, it doesn't know your product, audience or positioning unless you tell it. A project with your style guide, product facts and a few strong example posts fixes most of that (Anthropic).

Our view: the teams getting the most from Claude in SEO aren't using it to write more pages. They use it to make each page better researched and to keep existing pages current, which is where most content programmes fall behind.

#02How do you turn keyword data into briefs?

Export a keyword cluster from your SEO tool and give Claude the data and the job.

Attached: a keyword export for the topic [topic] with columns
query, volume, difficulty, current position, URL.
1. Group the queries by search intent (learn, compare, buy, fix).
2. For each group, suggest whether it needs its own page or
   belongs as a section of an existing page. Use the URL column
   to spot existing pages.
3. For the top 3 opportunities, write a brief: working title,
   the main question the page must answer in its first 2
   sentences, H2 questions, facts we need to source, internal
   links to include, and what would make our page better than
   what usually ranks.
Don't invent volumes; use only the numbers in the file.

The last line matters. Claude will work with the numbers you provide, but if you ask it to estimate volumes without data it may produce figures that sound plausible and aren't. Anthropic's guidance on reducing hallucinations recommends explicitly allowing Claude to say it doesn't know (Anthropic).

#03How should you use Research for content?

Research is available on paid plans and needs web search turned on. Anthropic says Claude "runs multiple searches that build on each other", exploring different angles of your question, and it can also use your connected tools (Anthropic). The result is a cited report rather than a quick answer.

For SEO, Research is good for:

• Understanding a topic properly before briefing a writer, including official definitions, standards and recent changes.
• Collecting primary sources you can cite: documentation, regulations, original studies.
• Checking what's changed since an old post was written.

Research [topic] as of today. I need:
- The current official definition and who publishes it
- What changed in the last 12 months, with dates
- Primary sources (official docs, regulators, original research),
  not other blog posts, where possible
- Claims that commonly appear online but aren't supported by a
  primary source
Cite every claim. Flag anything you couldn't verify.

Then open the sources yourself. Research makes finding sources faster; it doesn't make checking them optional.

#04How do you refresh old content with Claude?

Content refreshes are often the best return on effort, and Claude is good at the comparison work.

Here's our post [paste or attach] and the current Search Console
data for its URL [paste queries, clicks, impressions, position].
1. List the queries we get impressions for but the post doesn't
   answer directly.
2. List statements that may be out of date; use web search to
   check each and cite what you find.
3. Suggest new H2 questions, a revised first paragraph that
   answers the main query directly, and sections to cut.
Keep our voice. Don't rewrite the whole post.

Asking for a list of changes rather than a full rewrite keeps you in control and makes review faster. It also preserves what's already working on the page.

#05Can Claude help with internal linking?

Yes, if you give it a list of your URLs and titles. Export your sitemap or a crawl, then:

Attached: our site's URLs with titles and H1s. For the draft
below, suggest 5-8 internal links: the exact anchor text from
the draft, the target URL, and why the link helps a reader.
Only use URLs from the list. Don't suggest links to pages that
compete for the same main query.

The "only use URLs from the list" instruction prevents invented URLs, which is a common problem when you ask for internal links without data.

#06How do you keep quality and voice consistent?

Set up a project for SEO content with your instructions and reference files. Anthropic notes that project knowledge is available across all chats in the project. Good things to include:

• Your style guide and banned phrases
• Product facts and pricing, with a date
• Two or three of your best-performing posts as examples
• Your brief template and editorial checklist

You can go further with a skill: a saved set of instructions and files Claude uses when relevant (Anthropic). A "content brief" skill that always produces your format saves pasting the template each time.

#07What can scheduled tasks do for SEO?

Scheduled tasks are available on all paid plans in Cowork. Anthropic says they run remotely, so they run "even when your computer is asleep", and they can use your connectors, skills and plugins (Anthropic). Useful SEO routines include:

• A weekly summary of new queries from a Search Console export saved to a shared folder
• A monthly check of your top pages for out-of-date statements, using web search
• A weekly digest of official search engine announcements with links

Keep scheduled tasks producing reports for a human to act on. Don't have them publish or edit live pages.

#08From MentionX data: whether SEO content also earns AI citationsOriginal data

Method. Read-only aggregate queries over MentionX scans from 22 August to 29 October 2026 (UTC): 2,217 engine runs (2,154 completed) across 18 anonymised projects and six engines.

Ranking in search and being cited by AI engines overlap but aren't the same. MentionX's aggregates show how often pages that rank also get cited by the engines we track:

Our view: briefs that start with a direct answer and use sourced facts tend to serve both search and AI engines. Add "answer the main question in the first two sentences" to every brief.

Limits. A convenience sample of mostly B2B software projects, run through engine APIs, which can behave differently from the consumer apps. Perplexity, Copilot and DeepSeek figures come from a single day (26 October 2026), so treat them as a snapshot, not a trend.

#09Worked example: Ledgerline's refresh sprint

Ledgerline is a fictional accounts-payable tool. Figures are illustrative.

Ledgerline's content lead picks 15 older posts with falling clicks. For each, she gives Claude the post and the Search Console queries and asks for a list of changes with cited updates.

Claude flags outdated tax deadlines in four posts, missing sections for queries with impressions in nine, and two posts that compete for the same query. She merges those two and has writers apply the rest, checking every cited source.

A monthly scheduled task now lists statements in the top 20 posts that may need checking. Ledgerline is fictional and the details are illustrative.

#10What should you try before the next lesson?

Take one older post with declining traffic. Export its Search Console queries, run the refresh prompt and apply only the changes you can verify. Compare the result in a month.

Next: Ad copy and testing with Claude. Or return to the Claude for Marketers track.

#11What does Google say about AI-assisted content?

Google's guidance on generative AI content focuses on the result, not the tool. It says using automation, including AI, to generate content with the primary purpose of manipulating search rankings is a violation of its spam policies, while AI can be helpful for creating content that is useful and original (Google Search Central). The practical reading for marketers is simple: Claude can help you produce better pages, but publishing large volumes of thin pages because they're cheap to make is exactly the behaviour the policy targets.

That's another reason to point Claude at research, refreshes and editing rather than raw page count. A page that answers a question better than what already exists, with sourced facts and your own experience, is worth more than ten that repeat what's already ranking.

#12How do you add real expertise that Claude can't supply?

Claude can organise and explain, but it can't supply your first-hand experience: what your customers actually ask, what happened when you tried something, the numbers from your own product. Those details are what make a page worth citing, by people and by AI engines.

Build that into the workflow rather than hoping writers remember it. Two prompts help.

Before drafting, ask Claude to interview the subject-matter expert through you: "Based on this brief, write eight questions to ask our product specialist that would give us details no competitor could write." Send the questions, paste the answers back, and have Claude work them into the draft.

After drafting, ask Claude to mark the gaps: "Highlight every paragraph that could appear on any competitor's site unchanged. For each, suggest what specific detail from our experience would make it ours." The highlighted paragraphs are where an editor's time is best spent.

#13How do you check a draft before it goes live?

Use Claude as a second editor with a fixed checklist, ideally in a fresh chat so it isn't marking its own work from memory of writing it.

Ask it to check that the first two sentences answer the page's main question, that each H2 is a question a reader would actually ask, that every statistic has a linked source, that product facts match your reference file and that the meta description is within your length limit. Ask for pass or fail with a quote as evidence for each item. Anthropic's hallucination guidance suggests asking Claude to find a supporting quote for each claim and to retract any it can't support, which works well as a final pass on facts.

Then a human reads it. The checklist catches the routine problems; an editor catches the ones that need judgement, like whether the advice is actually right for your audience.

#14How should a small team split the work?

For a team of one or two, a sensible division is: the human picks topics from data, gathers expert input and makes final edits; Claude handles clustering, briefs, first-draft structure, refresh lists and checklist reviews. For larger teams, put the brief template, checklist and product facts in a shared project or skill so every writer starts from the same instructions, and keep a short log of prompts that worked well.

Whatever the size, measure the outcome, not the output. Track clicks, conversions and citations for pages Claude helped with alongside the rest, and keep using the parts of the workflow that move those numbers.

#15What about structured data and technical SEO?

Claude can help here too, as long as you validate its output. It can draft FAQ or article schema from a finished page, explain what a crawl report's warnings mean, write regex for redirect rules or turn a messy list of old URLs into a redirect map. Treat each of these as a draft. Run schema through Google's Rich Results Test, test redirects on a staging site and have whoever owns the website review anything that touches templates or server settings.

The same goes for content audits at scale. Claude can sort a crawl export of several thousand URLs into keep, merge, refresh and remove groups based on rules you give it, such as traffic thresholds and topic overlap. That saves days of spreadsheet work. But the decisions to remove or redirect pages affect rankings, so a person should review the merge and remove lists before anything changes on the live site.

#16What does a complete content brief template look like?

Give Claude this template as a file in your SEO project and ask it to fill it in from your keyword data and research. A writer, human or Claude, should be able to produce a good draft from it without further questions.

Working title:
Primary query and close variants (from data):
Search intent and reader situation:
The answer in 2 sentences (goes at the top of the page):
H2 questions, in order:
For each H2: the point to make and the source to cite
Facts we must include (with sources):
Our first-hand angle (expert, data or customer story):
Internal links (from our URL list) and anchor text:
What the current top results miss:
Length guidance: as long as the questions need, no padding
Meta title (under 60 chars) and description (under 155):

Ask Claude to mark any field it can't fill from the data as [needs input], rather than guessing. Those gaps are your to-do list before writing.

#17How do you find keyword cannibalisation?

When several of your pages compete for the same query, they can split clicks and confuse search engines. Export Search Console data with query and page columns, then:

Attached: Search Console performance by query and page for
3 months. Find queries where 2 or more of our URLs get
meaningful impressions. For each, show the URLs with clicks,
impressions and average position, and suggest: merge,
differentiate (change one page's focus), or leave alone.
Show the threshold you used for "meaningful".

Review the suggestions with whoever owns the pages. Merges need redirects and sometimes content from both pages, so treat them as small projects.

#18What goes wrong in SEO workflows, and how do you fix it?

#19How do you run a quarterly content review?

Once a quarter, export page-level clicks and impressions for the last two quarters alongside a crawl with titles and publish dates. Ask Claude to sort pages into four groups: growing, stable, declining and never performed. For declining pages, run the refresh prompt. For pages that never performed, ask whether they target a real query and whether another page does it better. That last group is often where merging or removing pages gives the biggest improvement to the site as a whole.

#20Common questions

Can Claude do keyword research? It can group and analyse keyword data you provide and suggest topics, but it doesn't have search volume or ranking data. Use your SEO tools for that.

Is AI-written content bad for SEO? Google's guidance focuses on helpful, reliable content regardless of how it's produced. Quality, accuracy and original value matter more than the tool.

Can Claude publish to my CMS? Possibly through a connector or browser, but we recommend keeping publishing as a human step and using Claude for drafts and reports.